***** To join INSNA, visit http://www.insna.org ***** Dear Jenine, You have your nodes in two groups. It seems you want to partition your collection of links into two networks - those linking to pairs in the same classes and those to pairs in different classes. I think of this as two separate calculations or transformations. Assume that the original network is called is_linked_to, then we want to calculate two new networks: is_similar_to and is_dissimilar_to. All three networks may have the same nodes and is_similar_to links a->a and b->b whereas is_dissimilar_to links a->b and b->a. This can be done in VisuaLyzer of the SocioMetrica Suite at http://mdlogix.com/visualyzer.htm in a somewhat straightforward manner if your data can be exported from Pajek to UCINET format and then imported into VisuaLyzer. If your nodes have two attributes, a name and a class, then your partition is based on this second attribute. In the reasoning language this second attribute is selected with the relation, node^2-2. This effectively says 'Of the two attributes of the node, give me the second.' (Similarly, node^2-1 returns the name.) Alternatively, converse(node^2-2), constructs the second attribute. The composition of the two (the relative product) is (node^2-2:converse(node^2-2)); it is smaller than the identity relation. Therefore we can intersect this with the original to define is_similar_to as the product is_linked_to*(node^2-2:converse(node^2-2)). Procedurally, this is computed in VisuaLyzer by first typing this into the box called 'Relation Expression' brought to the top with the ^R command. Be sure to rename the resulting expression as is_similar_to and check the box labelled 'Extend the existing collection with this new network'. Also, we may define is_dissimilar_to as is_linked_to*(node^2-2:di:converse(node^2-2)) since the apartness relation is di. All this results in three networks linking the same set of nodes. Some understanding may come by observing that is_linked_to is the sum (union) of the other two collections of links. I hope this helps. This may require some experimentation, a fundamentally different way of thinking, and more discussion. paul jenine harris wrote: > ***** To join INSNA, visit http://www.insna.org ***** > > Hi networkers, > > I have a large-ish directed network (1877 nodes) in Pajek with a > partition > that classifies each node into one of two categories. For the most > part the > nodes in each category only are connected to other nodes in the same > category (Category A --> Category A). However, there is a proportion of > nodes that are linked to nodes in the other category (Category A --> > Category B). Is there a way in Pajek to extract/identify these nodes that > are involved in cross-category connections? I've gone through the book > and > have moderate experience with the software and still just can't come > up with > anything. My final goal here is to make a new partition with Category A, > Category B, and Category AB which would be those nodes involved in cross > category connections. > > I also have access to UCINET, but am much more comfortable with Pajek. > > Thanks in advance for your help. > > See you in Corfu! > > jenine > _____________________________________________________________________ SOCNET is a service of INSNA, the professional association for social network researchers (http://www.insna.org). To unsubscribe, send an email message to [log in to unmask] containing the line UNSUBSCRIBE SOCNET in the body of the message.